[{"data":1,"prerenderedAt":27},["ShallowReactive",2],{"chapter:vision-rs\u002Fintroduction.json":3},{"project":4,"route":5,"title":6,"titleHtml":6,"navTitle":6,"part":7,"sourcePath":8,"editUrl":9,"html":10,"toc":11,"hasMermaid":22,"prev":7,"next":23},"vision-rs","\u002Fvision-rs","Introduction",null,"introduction.md","https:\u002F\u002Fgithub.com\u002Fteenygrad\u002Fvision-rs\u002Fedit\u002Fmain\u002Fbook\u002Fsrc\u002Fintroduction.md","\u003Cp>\u003Cstrong>vision-rs\u003C\u002Fstrong> is a high-performance computer vision SDK for Rust, built on\n\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fteenygrad\u002Fteenygrad\" target=\"_blank\" rel=\"noopener noreferrer\">teenygrad\u003C\u002Fa>, a memory-safe Rust ML\ntraining and inference runtime. It currently ships \u003Cstrong>YOLO26\u003C\u002Fstrong>, a real-time\nobject detection model family, with custom GPU kernels (attention, detection\ndecode, loss functions) written directly against teenygrad’s Triton-style\nkernel DSL.\u003C\u002Fp>\n\u003Ch2 id=\"what-this-book-covers\">What this book covers\u003C\u002Fh2>\n\u003Cul>\n\u003Cli>Getting a working detector running against a JPEG\u002FPNG image.\u003C\u002Fli>\n\u003Cli>The \u003Ccode>ObjectDetector\u003C\u002Fcode>\u002F\u003Ccode>DetectorConfig\u003C\u002Fcode> API surface.\u003C\u002Fli>\n\u003Cli>How the YOLO26 model is assembled from backbone\u002Fneck\u002Fhead blocks, and how\nthe N\u002FS\u002FM\u002FL\u002FXL variants differ.\u003C\u002Fli>\n\u003Cli>Training: the TaskAlignedAssigner, CIoU\u002Fclassification losses, and the\ndual-head (one2many\u002Fone2one) assignment scheme.\u003C\u002Fli>\n\u003Cli>The custom kernels vision-rs ships (Flash Attention 2, position-sensitive\nattention, detect-decode) and the \u003Ccode>teenyc\u003C\u002Fcode> toolchain that compiles them.\u003C\u002Fli>\n\u003Cli>Cross-compiling and packaging a self-contained deployable bundle for a\nJetson Orin Nano.\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch2 id=\"what-this-book-does-not-cover\">What this book does not cover\u003C\u002Fh2>\n\u003Cp>This book is a guide to \u003Cem>using\u003C\u002Fem> vision-rs. For the full API reference\n(every public type, field, and function), see the\n\u003Ca href=\"\u002Fapi\u002Fvision-rs\u002Fvision_rs\u002Findex.html\">crate documentation\u003C\u002Fa>. For the ML runtime\nvision-rs is built on — tensors, the computational graph, the kernel\ncompiler — see \u003Ca href=\"\u002Fteenygrad\">The Teenygrad Book\u003C\u002Fa>.\u003C\u002Fp>\n\u003Ch2 id=\"status\">Status\u003C\u002Fh2>\n\u003Cp>vision-rs is under active development. The public API (\u003Ccode>vision_rs::detect\u003C\u002Fcode>)\nis the intended stable surface; everything under \u003Ccode>vision_rs::models\u003C\u002Fcode> is\nexpected to grow additional model families over time. See the\n\u003Ca href=\"\u002Fvision-rs\u002Fappendix\u002Ffaq-and-roadmap\">FAQ & Roadmap\u003C\u002Fa> for what’s implemented today.\u003C\u002Fp>\n",[12,16,19],{"id":13,"text":14,"level":15},"what-this-book-covers","What this book covers",2,{"id":17,"text":18,"level":15},"what-this-book-does-not-cover","What this book does not cover",{"id":20,"text":21,"level":15},"status","Status",false,{"title":24,"titleHtml":25,"route":26},"Installation & Toolchain","Installation & Toolchain","\u002Fvision-rs\u002Fgetting-started\u002Finstallation",1786271830393]